0.50 - added another error text to detect memory overrun failure
0.49 - added another error text to detect memory overrun failure
0.48 - log the reason that triggered vmem retry testing
0.47 - adjust memory retry increment and usage reporting when multiple cpus
are used for one stage (assumes that the requested number were actually
provided - that might not match the configured environment definition;
but in our lab at least, the accounting info always returns undefined
value for the slots allocated)
0.46 - documented retry_mem_percent attribute, added another error text to
detect memory overrun failure
0.45 - changed the magic number 99 to a stage attribute that can be set to
allow causing retry with additional memory for a different threshold
when the 'Out of memofy!' message is not generated.
0.44 - fix DRMAARun - it was re-using a variable for every stage instead of
having an object private one that re-initialized to undef, leaking
values
0.43 - edit the diagnostic when drmaa_wait fails to get status info
- that seems to be a timing bug in the drmaa library interaction with
the sge code
0.42 - added stage name ad run index to warn message, in case running at
log_level 'WARN' the message needs to be self-sufficient
0.41 - add memory_too_small function ref attribute to allow a user-specified
method to decide that the job was killed for using too much memory (in
case the standard test misses some cases)
0.40 - fix testing of extra_sge_args_string to correctly detect collisions
with internally provided values
- fix test 07-extra.args.t to pass its args with the right name
0.39 - fix typo - _drmaa_job_id -> _drmaa_jobid
0.38 - update MANIFEST
0.37 - work around drmaa handling of -q arg differently from qsub
0.36 - removed some extraneous use MODULEs
- fallback to qacct if drmaa fails to get exit_status info
- added missing required modules to Build.PL
0.35 - map generic time specifiers (e.g. '3d2h10m') to native SGE permitted
form
0.34 - typo bugfixes
0.33 - added support for -pe to submit parallel jobs
0.32 - improved detection for job terminated because of using too much memory
- added support for abort for when drmaa says a job was not accepted
- added direct support for kill, killsignal rather than leaving them
tangled in exit_status and subject to local interpretation
- rearranged a number of methods and modifiers so that they could be
overridden in drivers smoothly
- removed modules_to_load(Perl-BL) from tests, they only work in BoutrosLab,
not elsewhere
- adjusted to make HPCI::ModuleLoad not automatically loaded, use a config
wrapper module for your lab if needed, similarily for HPCI::ScriptSource
- added support for using DRMAA to run jobs
TODO: move historical change info into here